UniMate: One Unified Model to Animate Diverse Skeletons
UniMate claims one text-driven animation model can move many skeleton types without per-skeleton retraining.
The paper targets a bottleneck left by automatic rigging: generating motion for newly rigged 3D assets. Its model uses skeletal topology inside a diffusion transformer, including graph-aware attention, spectral position embeddings, and a global topology conditioner. The authors also introduce UniML3D, a 13,006-sequence dataset covering bipeds, quadrupeds, birds, marine forms, insects, snakes, and articulated rigid objects. They report stronger quality, generalization, and efficiency than baselines, plus zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing. HF Daily Papers' note
The paper targets a bottleneck left by automatic rigging: generating motion for newly rigged 3D assets. Its model uses skeletal topology inside a diffusion transformer, including graph-aware attention, spectral position embeddings, and a global topology conditioner. The authors also introduce UniML3D, a 13,006-sequence dataset covering bipeds, quadrupeds, birds, marine forms, insects, snakes, and articulated rigid objects. They report stronger quality, generalization, and efficiency than baselines, plus zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing. HF Daily Papers' note
score 5